The spatial landscape of clonal somatic mutations in benign and malignant tissue Joakim Lundeberg ( [email protected]) KTH Royal Institute of Technology https://orcid.org/0000-0003-4313-1601 Andrew Erickson University of Oxford Emelie Berglund KTH Royal Institute of Technology Mengxiao He KTH Royal Institute of Technology https://orcid.org/0000-0001-5905-8467 Maja Marklund KTH Royal Institute of Technology Reza Mirzazadeh KTH Royal Institute of Technology Niklas Schultz Karolinska Institutet Linda Kvastad KTH Royal Institute of Technology Alma Andersson KTH Royal Institute of Technology Ludvig Bergenstråhle KTH Royal Institute of Technology Joseph Bergenstråhle KTH Royal Institute of Technology Ludvig Larsson Karolinska Institute https://orcid.org/0000-0003-4209-2911 Alia Shamikh Karolinska Institutet Elisa Basmaci Karolinska Institutet Teresita Diaz de Ståhl Karolinska Institutet Timothy Rajakumar University of Oxford Kim Thrane
40
Embed
The spatial landscape of clonal somatic mutations in ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
The spatial landscape of clonal somatic mutationsin benign and malignant tissueJoakim Lundeberg ( [email protected] )
KTH Royal Institute of Technology https://orcid.org/0000-0003-4313-1601Andrew Erickson
University of OxfordEmelie Berglund
KTH Royal Institute of TechnologyMengxiao He
KTH Royal Institute of Technology https://orcid.org/0000-0001-5905-8467Maja Marklund
KTH Royal Institute of TechnologyReza Mirzazadeh
KTH Royal Institute of TechnologyNiklas Schultz
Karolinska InstitutetLinda Kvastad
KTH Royal Institute of TechnologyAlma Andersson
KTH Royal Institute of TechnologyLudvig Bergenstråhle
KTH Royal Institute of TechnologyJoseph Bergenstråhle
KTH Royal Institute of TechnologyLudvig Larsson
Karolinska Institute https://orcid.org/0000-0003-4209-2911Alia Shamikh
The spatial landscape of clonal somatic mutations in benign and malignant
tissue
Andrew Erickson1,§, Emelie Berglund2,§, Mengxiao He2¤, Maja Marklund2,¤, Reza Mirzazadeh2¤,
Niklas Schultz3,¤, Linda Kvastad2, Alma Andersson2, Ludvig Bergenstråhle2, Joseph
Bergenstråhle2, Ludvig Larsson2, Alia Shamikh4,5, Elisa Basmaci4,5, Teresita Diaz De Ståhl4,5,
Timothy Rajakumar1, Kim Thrane2, Andrew L Ji6, Paul A Khavari6, Firaz Tarish3, Anna
Tanoglidi7, Jonas Maaskola2, Richard Colling1,8, Tuomas Mirtti9,10,11, Freddie C Hamdy1,12, Dan
J Woodcock1,13, Thomas Helleday3,14, Ian G Mills1, Alastair D Lamb1,12⇞ and Joakim
Lundeberg2,⇞*
1Nuffield Department of Surgical Sciences, University of Oxford, UK 2Department of Gene Technology, KTH Royal Institute of Technology, Science for Life
Laboratory, Solna, Sweden. 3Science for Life Laboratory, Department of Oncology-Pathology, Karolinska Institutet, Solna,
Sweden. 4Department of Oncology-Pathology, Karolinska Institutet, Stockholm, Sweden 5Department of Clinical Pathology and Cytology, Karolinska University Hospital, Stockholm,
Sweden 6Program in Epithelial Biology, Stanford University School of Medicine, Stanford, USA 7Department of Clinical Pathology, University Uppsala Hospital, Uppsala, Sweden. 8Department of Cellular Pathology, Oxford University Hospitals NHS Foundation Trust, UK 9Department of Pathology, University of Helsinki & Helsinki University Hospital, Finland 10Research Program in Systems Oncology, Faculty of Medicine, University of Helsinki, Finland. 11iCAN-Digital Precision Cancer Medicine Flagship, Helsinki, Finland. 12Department of Urology, Oxford University Hospitals NHS Foundation Trust, UK 14Weston Park Cancer Centre, Department of Oncology and Metabolism, University of Sheffield,
UK 13Big Data Institute, Old Road Campus, University of Oxford, UK
§These authors contributed equally to this work. ¤These authors contributed equally to this work. ⇞Joint senior authors. *Correspondence should be addressed to J.L. ([email protected])
1
Abstract 1
Defining the transition from benign to malignant tissue is fundamental to improve early 2
diagnosis of cancer. Here, we provide an unsupervised approach to study spatial genome 3
integrity in situ to describe previously unidentified clonal relationships. We employed spatially 4
resolved transcriptomics to infer spatial copy number variations in >120 000 regions across 5
multiple organs, in benign and malignant tissues. We demonstrate that genome-wide copy 6
number variation reveals distinct clonal patterns within tumours and in nearby benign tissue 7
using an organ-wide approach focused on the prostate. Our results suggest a model for how 8
genomic instability arises in histologically benign tissue that may represent early events in 9
cancer evolution. We highlight the power of capturing the molecular and spatial continuums in a 10
tissue context and challenge the rationale for treatment paradigms, including focal therapy. 11
2
Main 1
Mutations can either be inherited or acquired (somatic). Inherited genomic alterations are readily 2
identifiable as these are present in all cells while somatic mutations are usually only present in a 3
small fraction of cells. In order to obtain spatial information of these rarer non-heritable genetic 4
events occurring in cancer, studies have commonly used laser capture microdissection to retrieve 5
histologically (or biomarker) defined tissue regions or even single cells. These studies have an 6
inherent bias as only a limited number of spatial regions or single cells per tissue section can be 7
collected and examined. The possibility to perform spatial genome analysis without being 8
confined by histological boundaries would therefore provide an important contribution to 9
delineate the clonal architecture in tumours and co-existing benign tissue. 10
Spatially resolved transcriptomics has emerged as a genome-wide methodology to explore 11
tissues in an unsupervised manner1. In this study we infer genome-wide copy-number variations 12
(CNV) from spatially resolved mRNA profiles in situ (Fig. 1a). Gene expression has previously 13
been used to inferCNVs in single cells, successfully identifying regions of chromosomal (chr) 14
gain and loss2. Here we expand into a spatial modality generating CNV calls in each spatial 15
region represented by barcoded spots. First, we sought corroboration that inferCNV (iCNV) data 16
could accurately mirror DNA-based phylogenies, using simultaneously extracted single cell 17
RNA, and DNA3 (Extended Data Fig. 1a). Next, we successfully recapitulated published DNA-18
based phylogenies in prostate cancer using RNA from the same samples4–6 (Extended Data Fig. 19
1b, c). To ensure that we robustly could capture sufficient and accurate CNV information from 20
individual spots from a multifocal tumour model, and use this information to deduce clonal 21
relationships between cells, we then designed an in-silico system to synthesise a tissue 22
containing multiple clones determined by stochastic copy number (CN) mutations in a single 23
3
artificial chromosome. Using a probabilistic method to generate gene expression from such 1
mutations we then interrogated the expression data using iCNV, while blind to the underlying 2
‘ground-truth’ CN status, and successfully recapitulated both the CN status and the clonal 3
groupings (Extended Data Fig. 2a-c). 4
Next, we used a cross section of an entire prostate organ to explore the spatial iCNV landscape 5
of a commonly multifocal malignancy7. The specimen was obtained by open radical 6
prostatectomy and an axial section was taken from the mid-gland. The axial section was 7
subdivided into cubes (Fig. 1a, b) and corresponding tissue sections were histologically graded 8
using the Gleason grading system8 identifying extensive intratumoural heterogeneity (ITH) in the 9
context of surrounding benign tissue (Fig. 1b, e). We obtained organ-wide transcriptional 10
information from 21 cubes (tissue sections) and > 21 000 barcoded regions (100 micron spots) 11
with a mean average of 3500 expressed genes detected per barcoded spot9. We then analysed the 12
transcriptional data using factorized negative binomial regression (Extended Data Fig. 3a). This 13
provided an unsupervised view of gene expression factors (GEFs) over the cross section of the 14
prostate (Fig. 1c). Twenty-five factors showed overlap between histology and GEFs representing 15
tumour, hyperplasia and benign epithelia annotated by the factor marker genes, as previously 16
reported10 (Fig. 1f). Several GEFs provided distinct ‘clonal’ appearances and were associated 17
with tumour regions (Fig. 1f, right panel). Next, we undertook a spatial iCNV analysis to provide 18
an overall landscape of genome integrity (Fig. 1d) identifying certain regions with increased 19
iCNV activity (V1_1, H2_1, H1_1, H1_5, H2_5; Fig. 1g) while the majority of the tissue area 20
appears to be CN neutral. These initial results suggested that iCNVs could identify tissue 21
regions, at organ scale, with inferred genomic variability, distinct from morphology or 22
expression analysis. 23
4
To increase the fidelity of our analysis of variable iCNV regions we took advantage of smaller 1
55 micron diameter barcoded spots (Visium, 10x Genomics), reducing the number of cells to 2
approximately 5-10 per spot, to perform a more detailed interrogation of seven key sections. We 3
first validated the increased precision of this higher resolution platform using the synthetic tissue 4
method (Extended Data Fig. 2d, e). We next obtained data from approximately 30 000 spots 5
using factorized negative binomial regression resulting in 24 spatially distinct GEFs (Extended 6
data Fig. 3b). Two pathologists independently annotated each spot to provide consensus 7
pathology and histology scoring (Fig. 1e). We then investigated clonal relationships across the 8
investigated tissue using iCNVs. Having established the association between gene expression 9
factors and certain regions of interest (Fig. 1c, f) we wanted to determine the degree of clonal 10
CN heterogeneity in these regions. After designating all histologically benign spots as a 11
reference set (Extended Data Fig. 3c) it was immediately apparent that while certain GEFs 12
displayed a fairly homogenous inferred genotype (e.g. GEF 7, 14 and 22, Extended Data Fig. 13
3d), others were strikingly heterogeneous (e.g. GEF 10, Extended Data Fig. 3e). 14
Prompted by the realization that certain regions annotated as histologically benign displayed CN 15
heterogeneity (Fig. 1d), we refined the reference set to those spots which were both 16
histologically benign (outside the regions of interest) and also lacking any iCNV (Extended Data 17
Fig. 4). This constituted a ‘pure benign’ reference set for all subsequent iCNV analyses, unique 18
to each patient. It was apparent from cancer-wide inferred genotype (Fig. 2a-e) that there was 19
evidence of clonally distributed CN heterogeneity within areas of spatially homogeneous 20
Gleason pattern (Fig. 2a, d, e). We constructed a clone-tree to describe sequential clonal events 21
versus independently arising cancer-clones (Fig. 2b). It was apparent that two cancer clones 22
lacked key truncal events including a loss of a region of chr 16q and 8p which were otherwise 23
5
ubiquitous across all cancer clones (clones A and B, Fig. 2a, b). These clones were spatially 1
restricted to section H1_2 containing a region of low-grade Gleason Grade Group 1, discussed 2
later. The majority of clonally related spots were located around the largest focus of Gleason 3
Grade Group 4 disease with a striking pattern of truncal and branching events (clones H, I, J and 4
K). We therefore focused on this dominant region of cancer (spanning sections H1_4, H1_5 and 5
H2_5), to establish a first view of the interplay between spatial architecture and clonal dynamics 6
(remaining sections in Extended data Fig. 5a, b). 7
To construct clone-trees, we assumed that: (i) groups of cells containing identical CN profiles 8
were more likely to be related, than to have arisen by chance; and (ii) somatic CN events must be 9
acquired sequentially over time, the more numerous the events, the more distinct the clone. 10
Using this approach, we observed a common ancestral clone (clone H, Fig. 2b) containing 11
truncal events including CN loss on chr 6q and 16q, and CN gain on 12q and 16q. These were 12
clearly located in two tissue regions: an area of Gleason Grade Group 2 on the medial side of the 13
main tumour focus (section H1_4) and a region described as ‘transition state’ by consensus 14
pathology at the upper mid edge (section H2_5). These conserved iCNV features in distinct 15
spatial locations are noteworthy. A possible explanation is that clone H represents a linear 16
sequence of branching morphology in the prostatic glandular system11, and that further somatic 17
events took place giving rise to clones I, J and K forming a high-grade tumour focus (Fig. 2b), 18
which pushed apart the branching histology due to an aggressive expansile phenotype. For the 19
first time we have a spatial imprint of these events in prostate tissue. We also propose that some 20
CNVs may be of particular pathological significance (Extended data Fig. 4d) based on spatial 21
molecular phylogeny. Our analysis therefore provides insight into processes of tumour clonal 22
evolution, identifying discriminating events by spot-level CNV calling in a spatial context. 23
6
Given this discovery of a discordance between cellular phenotype and inferred genotype, we then 1
undertook a detailed interrogation of section H2_1 in the left peripheral zone of the prostate (Fig. 2
1c, 2c) containing roughly equal proportions of cancer and benign tissue. We profiled CN status 3
of every spot in this section and ordered these spots by hierarchical clustering into ‘clones’ A to 4
G based on defined levels of cluster separation (Fig. 3a, b). Spatially, we observed that these 5
data-driven ‘clone’ clusters were located in groups, broadly correlating with histological subtype, 6
but with some important distinctions (Fig. 3c, d). We observed that many CNVs already 7
occurred in benign tissue, clone C (Fig. 3a-d), most notably in chr 8, which has been well-8
described in aggressive prostate cancer12–14, but also several other CN gains and losses. Spatially, 9
this clone constituted a region of exclusively benign acinar cells branching off a duct lined by 10
largely copy neutral cells in nearby clone A and B (Fig. 3d). The unobserved ancestor to clone C 11
then gave rise to a further unobserved clone, and then cancer-containing clones E, F and G. 12
While clone G was made up exclusively of Gleason Grade Group 2 cancer cells, clones E and F 13
were mixed with up to 25% benign cells (Fig. 3d). The presence of somatic events in histological 14
benign cells highlights, for the first time in prostate biology, that these clone groups traverse 15
histological boundaries. 16
In order to validate that this inferred CN status was truly representative of underlying genotype 17
we used fluorescence in situ hybridization (FISH) probes to target two specific genes of 18
discriminatory interest, MYC and PTEN, encompassed in the notable chromosomal changes in 19
benign tissue clone C as well as high grade tumour clones, but absent in low grade disease. This 20
confirmed that while the status of both genes was diploid in normal benign tissue (Clone A), 21
MYC amplification and PTEN loss were evident in altered benign (Clone C) as well as in tumour 22
clones (Clone F; Fig. 3e, Extended data Fig. 6). Going forward, we hypothesise that other 23
7
homogenous iCNV calls are accurate, based on the evidence provided by these two selected loci. 1
This evidence suggests that somatic events, creating a mosaic of branching clones during ductal 2
morphogenesis, are present even in histologically benign disease. It therefore follows that an 3
understanding of this somatic mosaicism could distinguish which regions of benign glandular 4
tissue may give rise to lethal cancer, and which will not. 5
We considered the place of branching morphogenesis in the sequential acquisition of 6
transformative events in a predominantly benign section of the prostate (section H2_1, also in 7
section H2_2) (Extended data Fig. 7). Here we noted that such events seem to occur during the 8
development of prostatic ducts and acinar branches, with changes occurring at key branching 9
points, and the altered genotype passed on to daughter cells lining the ducts and glands of 10
associated branches. Interestingly, not all cells in such branches displayed the same cellular 11
structure, raising important questions as to why epithelial glands with seemingly identical 12